Research On Short Term Load Forecasting Method Of Power System Based On Generation Countermeasure Network

With the development of science and technology,the level of electrical automation continues to improve,more and more power generation forms of power access to the grid,followed by its load demand is also changing.Power system load forecasting refers to the prediction of the power demand and load fluctuation in the future.Accurate short-term load forecasting provides an important reference for the power system to make plans and schemes,at the same time,it is of great significance to avoid waste of resources,ensure the safe and reliable operation of power grid,and improve economic benefits.Considering that the general regression model is mainly built for linear relationship,ignoring the influence of climate,date type and other factors on short-term load forecasting,thus reducing the accuracy of forecasting.In order to learn the deep relationship implied in nonlinear load data and consider the influence of climate,date type and other factors on short-term load forecasting,this paper proposes a short-term load forecasting method based on generation antagonism network.In this method,convolution neural network is used to construct generation model and discriminant model.Historical load data,climate,date type and other load influencing factors data are used as condition and noise input generation model.Condition data,generated samples and real samples are used as input data to discriminant model respectively,In order to improve the forecasting accuracy,a feature loss function is introduced into the hidden layer of the discriminant model.In the practical application of load forecasting,the noise and load influencing factors data can be input into the trained model to forecast.In this paper,the three-year load of an area in the United States is taken as an example to compare the proposed method with the traditional load forecasting model and other deep learning algorithm forecasting models.Through a specific example,it is verified that the proposed method has outstanding advantages in prediction accuracy.